A black-and-white photograph of an object or a geographic area is a two dimensional construct of the actual image or area—for each x, y coordinate in the image there is a single value blackness or whiteness of that particular image spot. As human beings, the eye can perceive useful information about objects or areas based on the differences between black, white, and the shades of gray.
Color photographs add more visual information, but for most purposes the color information that is represented is tied to the visual spectrum. For each x, y coordinate in the image there is an approximation of the visual color spectrum of that particular image spot created through the blending of three color values, such as for example Red, Green, and Blue.
Multi-spectral sensing systems such as the Landsat Thematic Mapper remote imager and weather satellites produce images with a few relatively broad wavelength bands. The imager may capture a visual spectrum image and also one in infrared, but still they are limited in their ability to perceive information that may otherwise be present in a different part of the spectrum.
Hyperspectral sensors, on the other hand, collect image data across dozens if not hundreds of spectral bands, combining the technology of spectroscopy and remote imaging. The measurements captured by hyperspectral sensors make it possible to derive a contiguous spectrum for each image pixel. In other words for each x, y coordinate of an image (i.e., a pixel), rather than a single value for a gray or visible color, there is a third dimension—a vector, providing distinct information for that particular pixel across the large spectrum of wavelengths.
As different materials reflect wavelengths of visible and invisible light selectively, analysis of the contiguous wavelength spectrum permits finer resolution and greater perception of information contained in the image, through separation and evaluation of different wavelengths. For example, inorganic materials such as minerals, chemical compositions and crystalline structures control the shape of a representative spectral curve and the presence and positions of specific absorption bands.
The spectral reflectance curves of organic materials, such as healthy green plants also have a characteristic shape that is dictated by various plant attributes, such as the absorption effects from chlorophyll and other leaf pigments. Leaf structure varies significantly between plant species, and can be affected by plant stress. Therefore species type, plant stress and canopy state can all affect near infrared reflectance measurements, which are captured by hyperspectral sensors.
In addition, for a given pixel, a combination of different materials, e.g., biological, chemical, mineral, will provide a composite signal. Upon analysis and through comparison to known signal waveforms (e.g., known spectra) it is frequently possible to derive the presence of materials within a pixel, and therefore appreciate a detection granularity that is greater than the actual pixel resolution.
Hyperspectral sensors providing hyperspectral imaging can therefore be beneficially applied in a wide array of practical applications. Examples of such uses include aid in the detection of chemical or biological weapons, bomb damage assessment of underground structures, drug production and cultivation, as well as the detection of friend or foe troops and vehicles beneath foliage or camouflage.
Some targets are relatively easy to detect using standard techniques; whereas, other may not be. For example, detection of a terrain, such as asphalt, or concrete may be relatively straightforward for some images in which pixels (ground resolution cells) are filled by substantially the same material (e.g., asphalt or concrete). Alternatively, the measured signatures of a dispersed target, such as a gaseous plume, are complicated by a combination of phenomenology including effects of the atmosphere, spectral characteristics of the background material under the plume, temperature contrast between the gas and the surface, and the concentration of the gas. All of these quantities vary spatially further complicating the detection problem. For example, an effluent target in a low wind and having a relatively dense and very bright target signature on at least a few contiguous pixels may be relatively easy to detect, even with a substantially high threshold. Accordingly, such relatively easy to detect targets would require minimal or no spatial processing. Alternatively, targets in a high wind and/or sparse, or weak may be present in dozens to hundreds pixels of a given image. Unfortunately, all or most such pixels may be below conventional thresholds.
Spectral references of targets in a hyperspectral image scene are compared to each pixel to determine the presence/absence of the targets in that pixel. Detection of such dispersed targets requires integration of signal along plume by algorithm or eye (e.g., an analyst) without confusion from noise, clutter, and/or sensor artifacts.
One approach for improving detection would be to enhance the resolution of the remote sensing system used in obtaining a hyperspectral image (HSI). It is generally understood that a minimum detectable quantity (MDQ) scales with resolution of the imaged terrain, referred to herein as ground sample distance (GSD). A small GSD associated with a high resolution would improve detectability of such targets as each sampled cell would have a higher percentage of the target. Unfortunately, a small GSD would be an expensive quantity to achieve. This would be particularly true for HSI data obtained from space-based sensors.
Widely used target identification techniques include Spectral Matched Filter (SMF), also called Constrained Energy Minimization, and Adaptive Coherence/Cosine Estimator (ACE). Each of these techniques compares a priori known target spectrum to measurement (e.g., pixel spectra of an HSI data cube). Such techniques may be modified by an estimate of the image background. A common family of algorithms for statistically characterizing a scene's background use a background covariance matrix Σ. Both algorithms are computationally efficient and easy to implement and detect fairly bright targets in clutter. Unfortunately, both produced relatively high levels of false alarms, for dispersed targets, such as dim, gaseous effluents. This is due at least in part to targets having lower filter scores than many background pixels.
Other techniques addressing this problem describe mixture tuned matched filters, for example, also using minimum noise fraction transforms, such as the work of Mundt, J., Streutker, D. and Glenn, N. in “Partial Unmixing of Hyperspectral Imagery: Theory and Methods,” Proceedings of the American Society of Photogrammetry and Remote Sensing (2007). Unfortunately, such alternatives are very complicated, requiring the use of minimum noise fraction transforms, interpolation of eigenvalue vectors and the like. Other approaches, such as the work of Lentilucci and Schott, in “Target Detection in a Structured Background Environment Using an Infeasibility Metric in an Invariant Space,” in Algorithms and Technologies for Multispectral Hyperspectral, and Ultraspectral Imagery XI, Shen and Lewis, Eds., Proceedings of SPIE Vol. 5806 are likewise computationally complex, also requiring spectral matched filtering, linear unmixing and minimum noise fraction techniques in addition to identification of a target covariance matrix.
Hence there is a need for an improvement in target detection techniques for use with hyperspectral images that overcome one or more of technical problems noted above.